Spatial Collectiive

Classification in Remote Sensing

Precious Bassey's photo
Precious Bassey
·

3 min read

Image Classification Techniques in Remote Sensing [Infographic] - GIS开发者

Overview

Classification in remote sensing involves categorizing pixels in an image into meaningful classes or themes, such as land cover types. This process is critical in environmental monitoring, urban planning, and resource management. In this article we will explore the two different methods of classification namely the Unsupervised and Supervised classifications, the steps to achieving them in ArcMap.

Data Source

Unsupervised Classification:

Is a machine learning technique that identifies patterns and groups in data without predefined labels, automatically discovering hidden structures and relationships within the dataset by clustering similar data points together. In simpler terms it’s giving a child a basket of mixed fruits without telling them their names. The child starts sorting them based on similarities they notice - maybe grouping round fruits together or separating green from red fruits. To make it more technical imagine giving a computer explore an image without prior instructions. It automatically groups pixels that look similar to each other based on their spectral characteristics. The computer finds natural patterns and clusters in the image on its own, which you then interpret afterwards.

Steps to carry out Unsupervised classification

  • Add your Landsat 8 images to the ArcMap canvas.

  • In the Arc toolbox > Data management tools > Raster > Raster Processing > Composite bands. The Composite bands is used to categorize the raster images, using the Band combination parameters for nature which in this case is (4, 3, 2, 8). Find reference in Datasets. Click on OK.

  • Add the shapefile of the location used in the Landsat 8 image. Using the “Select by Polygon” mark out the part of the shapefile you want to work on. Export the selected feature.

  • In the Arc toolbox > Data management tools > Raster > Raster Processing > Clip. Clip the area of interest.

  • Load and preprocess the Clipped image. For preprocessing in the Arc toolbox > Spatial Analyst Tools > Multivariate > Principal Components

  • Run Unsupervised classification analysis. For the analysis, Arc toolbox > Spatial Analyst Tools > Multivariate > Iso cluster Unsupervised Classification. Specify the number of desired classes, Click OK.

  • Interpret the result of the analysis by assigning meaningful labels to the clusters by comparing with reference data, this comparison is carried out using visual inspection.

  • Result of the Unsupervised Classification

Supervised Classifications

Is a machine learning technique where an algorithm learns to classify data by training on a labeled dataset, using predefined categories and known examples to predict the class of new, unseen data points. In simpler terms imagine you’re teaching a computer to recognize different types of land covers by showing it examples, you manually select and label specific areas on a satellite image - like “this is water”, “this is urban area” etc. The computer learns these characteristics and applies and applies that knowledge to classify the entire image, matching other pixels to these predefined categories.

Steps to carry out Supervised classification

Create training samples. The training samples are parts of the Landsat-8 images with different features as shown on the image below, this samples will be used to train the model on how to identify the different features.

Creating the training samples

  • Train the classifier with the classification algorithm, the Maximum Likelihood Classification. Arc toolbox > Spatial Analyst Tools > Multivariate > Maximum Likelihood Classification.

  • Result of the classification.

GISRemote Sensing classification